Weekend Reading #368

This is the three-hundredth-and-sixty-eighth weekly edition of our newsletter, Weekend Reading, sent out on Saturday 13th June 2026.

To receive a copy each week directly into your inbox, sign up here.

*****

What we're thinking.

Well today is the big day as SpaceX comes to market. The sheer scale of retail demand for this is quite staggering as apparently over $70 billion of pure retail demand has come through. Pretty much everyone we know has asked us what we think about the IPO and will it go up and as usual our answer is we don't know. While the following may be obvious it is worth repeating. Going in for an IPO is simply a bet that the opening price will be higher than the offer price. Once it is live, you can buy any time you want. So, it's always highly speculative. In this case you have the battle of gargantuan retail demand who are buying imagination and optionality versus institutions and so-called "sophisticated" market players who mostly appear to think it is obscenely expensive. In recent years, retail has tended to win this type of tug of war. It will be fun to observe what unfolds. Elsewhere another announcement from Trump about a deal caused a rally late this week after some serious selling. For the first time in a while we are in observer mode, waiting for the market to tell us what happens next.

 
What we're reading.

This anecdote about what happened with the introduction of electricity is a very good analogy to today's AI revolution. Originally when electricity was discovered, it took some time until factory processes and layouts were adjusted from first principles and initially most industries simply adopted a "faster horse" model. We still await the first AI native businesses in which entrepreneurs build AI first. Can you actually imagine what is coming?
 
In light of the above, here is a good clip from a former Anthropic developer who now uses Codex from Open AI as his preferred tool. The key takeaway from this for me is this. We will look back on how we manually search and key in information into web browsers with amusement once day. Our primary interface will be our AI agents. They will do the browsing etc. This means that websites will shift from needing an attractive user interface to being information hubs for our agents to interact with. Worth a watch. DC

Back in the early days of LLM excitement, an analogy was put to us of LLMs being a huge zip file of data, compressing the web into a fraction of its total size (albeit still big), and when prompted, an estimate of the most likely “correct” answer is produced. Fast forward 3 years or so and the flurry of “most likely correct” answers have now become the substrate on which new models are trained – the output from before is becoming the input. Net result: anything that isn’t “most likely correct” gets trimmed off, and everything tends towards the average central tendency. And no, this isn’t a case of making models with more parameters to fix the problem – it’s a feature of all regressions when trying to get the line of best fit. The title of this paper puts it best: “The curse of recursion: training on generated data makes model forget”. Because at the end of the day, it’s all the quirks, errors, character and personality of the writers that make content interesting, less so that it’s “correct”.
 
A short snippet recalling the events of 2012 spotted on twitter on Facebook’s IPO and its first day of trading – turns out the fumble at the beginning of Facebook’s first day of trading involved the now famous trading house Jane Street and a sensible but nonetheless wrong choice of variable type on a pricing system. Just another example of how really small things can have huge consequences if not managed properly. EL

What we're watching.

On the topic of SpaceX, this is a fantastic interview conducted by Gavin Baker with SpaceX CFO, Bret Johnsen.
 
Jeff Bezos is out of retirement, returning as co-CEO of his new AI startup, Prometheus. It reminds me of when Hulk Hogan came back from retirement. First he looks a lot more like him and second we have been starved of having him around in public for way too long. 
 
It's building what Bezos calls an "artificial general engineer". AI that designs jet engines, optimizes manufacturing, and prototypes physical products. LLMs learned from the internet's text. Prometheus is learning from the physical world physics, simulations, engineering data, manufacturing processes. In Bezos' own words: "Something that takes 100 engineers 10 years to build we want to make that 10 engineers, one year."
 
His key comment is that he expects job SHORTAGES from AI. 2nd week in a row of this. But my suspicion is that there may be a REDISTRIBUTION of jobs away from non-value add jobs (mainly white collar entry and mid level) towards PRODUCTIVE jobs. And because he believes there will be a productivity boom, Bezos is prescribing exactly that.
 
This clip of Microstrategy's Michael Saylor deserves a watch. In it he disdainfully claims he never ever said his company wouldn't sell bitcoin and that only "you" should. While on paper this may be true, this exchange is revealing. He is under serious pressure. Remember this is not his first rodeo. In the dot.com bust, he was a protagonist too. History rhymes etc. DC